1.3 Web Based Performance Model
Performance analysis using a web based performance model is executed in following three
phases as described below.
First, in the model construction phase, the objective of performance analysis is defined and the
model type is determined. Performance analysis can be used for various purposes such as
benchmarking, performance tests and capacity planning, etc… The objective needs to be clearly
defined in advance. Since there are two types of performance model, practical verification
method based on automated performance test solution and simulation verification method
based on analytical models, which method to use should be decided in advance. In general,
practical verification method is recommended for benchmarking and performance tests and
simulation verification method is recommended for capacity planning.
Second, in the model parameterization phase, the model parameter that determines the perfor-mance model is defined. For a web based system performance model, the two types of model
parameters are typically observed: Service Throughput which determines the workload and the
response/think time which describes the demand. The model parameter values may be
collected manually from the platform or systematically collected by using an APM (Application
Performance Monitoring) solution.
Finally, in the model validation phase, the actual parameter values are applied to the predefined
performance model for the verification purposes. In case of benchmarking or performance
tests, the user requests are simulated for system performance verification. In case of interpreta -tion models, a performance model based on the queuing theory is used to predict the future
trends of performance.
In a web based system, the service subject to analysis includes all dynamic contents (JSP /
Servlet / Web Service) executed by the web application servers, thus all request rate and the
response time for all the services must be known. In order to systematically measure of each
variable for large number of services, the gathering and analysis must be well planned out.
The correlations among the variables for this performance model are defined by the following
equations. Since details of the following equations are beyond the scope of this white paper;
please refer to the references attached at the end of this document.